GitLab

trunk: updates to scripts to more consistenly use new-style queue.pl options (--num-threads, --mem, --gpu) and remove --parallel-opts wherever possible (work out the new-style option from --num-threads directly where relevant)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@5104 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

Adding scripts incorporating word-dependent silence probabilities in making lexicon fst. Fixed a few bugs in some nnet2 scripts

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4880 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4830 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: committing accel2 version of train_pnorm script (basic p-norm training but with number of jobs that increases with time)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4824 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: Adding example script for 'multi-lingual' nnet2-online training setup, using RM and WSJ as examples.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4760 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4735 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: updates and fixes to some nnet2 training scripts, add 'accelerated' version of multisplice script.  No example scripts yet (still tuning).

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4717 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4673 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4622 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4620 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: nnet2 setup: Code and script changes to support dumping multi-frame egs to disk.  See get_egs2.sh, train_pnorm_simple2.sh, etc.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4614 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4550 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

several nnet2-online changes: make it easier to get the feature extraction options right in cross-system training; add train_pnorm_simple.sh script (simplified learning-rate schedule and improved combination at the end, supersedes train_pnorm_fast.sh); modifying big-data online-nnet2 recipes to use 40-dimensional MFCC rather than 13 as input (will add results soon, but they are improved).  Modified filter_scp.pl to have one-based, not zero-based, field index.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4493 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4409 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: removing --first-component-power option in p-norm training scripts (turned out not to be that useful).

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4394 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: merging from sandbox/online a couple files that were missed due to conflict (thanks Vassil for noticing)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4364 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: various mostly cosmetic changes to nnet scripts; add scripts in RM setup to demonstrate conversion nnet1->nnet2 followed by training.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4317 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

sandbox/online: demonstrating cross-system model transfer for online nnet2 decoding, see run_online_decoding_nnet2_wsj.sh

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/online@4278 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: committing changes to support 'perturbed' training (see train-nnet-perturbed.h).  This seems to only be subsantially helpful for small datasets like RM, and I may remove it in future.  Also adding a script that could be helpful for data cleanup, find_bad_utts.sh; this detects utterances where the transcript might not match with the audio.  Plus misc small fixes.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4241 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/online@4162 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/online@4160 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

sandbox/online: script and code updates intended to support training with online-estimated iVector; not tested yet.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/online@4159 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: update to 'fast' nnet training scripts to, on iterations where we add a layer, select the parallel training job that gives best training objective function, instead of just picking one arbitrarily.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4110 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: documentation improvements; add program that was missing from repo (est-pca); some updates to nnet2 recipes to use new, faster versions of the training scripts

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4089 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: modifying recipes for neural net training with the new online preconditioning; adding/modifying example scripts for RM and Fisher; various cosmetic code changes.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4088 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: changes to improve efficiency of new preconditioning method for nnet training; bug-fix in QR algorithm (for symmetric eigenvalue problem)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4080 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: modifications to recently-added new preconditioning method (to specify rank of input and output preconditioners separately)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4079 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4074 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: modify nnet2 training scripts to set priors based on posteriors on train data (not better but safer); some minor recipe fixes and updates, including raw-fMLLR+nnet training scripts.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3895 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: Script changes intended to make it harder to accidentally run GPU-optimized recipes when GPU support is not compiled into the code.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3873 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk/egs/babel/steps/nnet2: Modified get_lda.sh and train_pnorm.sh to optionally use FMLLR transforms from a directory different from the alignment directory

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3614 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/sandbox/language_id@3586 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: minor fix to scripts train_pnorm.sh and train_pnorn_ensemble.sh to work with non-GMM input model

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3553 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3491 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: small script changes (should have been checked in with bottleneck+DNN training setup for BABEL)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3465 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3442 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

Fix to discriminative training (retroactive thing and nnet-normalize-stddev), seems working, learning rate needs further tuning

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3421 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

trunk: change to nnet-combine-fast command line interpretation to make it easier to specify to start with averaged models; change to train_pnorm.sh script to give different options to nnet-combine-fast to make the final combination effective.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@3388 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8